Re: [problem] raid performance loss with 2.6.26-rc8 on 32-bit x86(bisected)

[Dan Williams]

On Wed, 2008-07-02 at 22:00 -0700, Mel Gorman wrote:

> Subject: [PATCH] Do not overwrite nr_zones on !NUMA when initialising zlcache_ptr
> 
> With the two-zonelist patches on !NUMA machines, there really is only one
> zonelist as __GFP_THISNODE is meaningless. However, during initialisation, the
> assumption is made that two zonelists exist when initialising zlcache_ptr. The
> result is that pgdat->nr_zones is always 0. As kswapd uses this value to
> determine what reclaim work is necessary, the result is that kswapd never
> reclaims. This causes processes to stall frequently in low-memory situations
> as they always direct reclaim.  This patch initialises zlcache_ptr correctly.
> 
> Signed-off-by: Mel Gorman <mel@xxxxxxxxx>
> ---
>  page_alloc.c |    1 -
>  1 file changed, 1 deletion(-)
> 
> diff -rup -X /usr/src/patchset-0.6/bin//dontdiff linux-2.6.26-rc8-clean/mm/page_alloc.c linux-2.6.26-rc8-fix-kswapd-on-numa/mm/page_alloc.c
> --- linux-2.6.26-rc8-clean/mm/page_alloc.c      2008-06-24 18:58:20.000000000 -0700
> +++ linux-2.6.26-rc8-fix-kswapd-on-numa/mm/page_alloc.c 2008-07-02 21:49:09.000000000 -0700
> @@ -2328,7 +2328,6 @@ static void build_zonelists(pg_data_t *p
>  static void build_zonelist_cache(pg_data_t *pgdat)
>  {
>         pgdat->node_zonelists[0].zlcache_ptr = NULL;
> -       pgdat->node_zonelists[1].zlcache_ptr = NULL;
>  }
> 
>  #endif /* CONFIG_NUMA */
> 

Bug squished.

# for i in `seq 1 5`; do dd if=/dev/zero of=/dev/md0 bs=1024k count=2048; done
2048+0 records in
2048+0 records out
2147483648 bytes (2.1 GB) copied, 7.73352 s, 278 MB/s
2048+0 records in
2048+0 records out
2147483648 bytes (2.1 GB) copied, 7.6845 s, 279 MB/s
2048+0 records in
2048+0 records out
2147483648 bytes (2.1 GB) copied, 7.74428 s, 277 MB/s
2048+0 records in
2048+0 records out
2147483648 bytes (2.1 GB) copied, 7.65959 s, 280 MB/s
2048+0 records in
2048+0 records out
2147483648 bytes (2.1 GB) copied, 7.73107 s, 278 MB/s

Tested-by: Dan Williams <dan.j.williams@xxxxxxxxx>



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/